Our take
AtlasCloud is a mid-size host for downloadable models with a strong tilt toward China-ecosystem offerings and competitive speed on smaller models. Its catalogue of 27 tracked offers includes several models rarely found at other hosts, though compliance documentation remains thin.
Use this for DeepSeek V4 Flash when roughly 45 tokens per second suffices. Pick it for high-throughput work on small models — Qwen3 6-35B A3B runs at 120 tps, well above the median. Choose it for China-ecosystem models such as Tencent, Meituan, Kwaipilot or MiniMax not widely hosted elsewhere. Skip it if you need verified headquarters location, EU endpoint, SOC 2, or zero-retention, or flagship throughput on larger models.
- Strong throughput on the smallest tracked model: Qwen3 6-35B A3B at 120 tps, 2.67× the next-highest throughput and 6.3× the catalogue median.
- Lowest input price among tracked offers: DeepSeek V4 Flash, 25% below the next-cheapest input tier.
- Does not train on prompts — privacy flag confirmed.
- Competitive throughput on mid-size models: Tencent HY3 at 66 tps and MiniMax M2.5 at 53 tps, both above the median of 44.5 tps for offers with throughput data.
- Compliance and geography documentation is thin: headquarters country, EU endpoint, SOC 2, and zero-retention option are all unverified in our data.
- Throughput on flagship models lags: DeepSeek V3-2 at 19 tps is 58% below the catalogue median of 45 tps for that model tier.
- Output prices escalate sharply on several models: Qwen3 6-35B A3B output is 5.99× its input price, and Meituan Longcat 2.0, MiniMax M2.5, and Kwaipilot Kat Coder Pro V2 all output at 4× their input price.
Privacy & data handling
- ✓
- Yes
- ✗
- No
- No answer on record
The colour says whether the answer favours you, not whether it is a yes — not training on your prompts earns a green cross, no zero-retention option earns a red one.